Research On Recognition Method For Helicopter Flight Action Based On Flight Data

With China’s gradual opening of low-altitude areas,it has witnessed the helicopter’susing in every aspect of people’s life.such as in disaster relief, forest planting, recyclingand so on.The more use of helicopter,the more helicopter pilots we need.As a result,how totraining a helicopter pilots faster and better become more and more important in thedepartment of military aviation’s research.Flight data recorded by this helicopter’s FDR can record all of the information duringthe flight,we have recollected of which can evaluate the flight actions totally. The methodcan make a great progress in evaluating the training quality, make it possible to superviseand improve the training quality scientifically, as well as provide technical ways forensuring flight security.In this dissertation, a helicopter’s six degrees of freedom flight dynamics model andhelicopter handling characteristics is introduced to find out the characteristic propertiesflight data who can reflect flight actions.For the data we selected has the phenomenon ofoutliers,noise,and data frame loss.We use wavelet transform method to locate the outliersand eliminate them.Then we contrast the results of the fast Fourier transform method andthe wavelet transform method for the noise reduction.Consequently,we fill and smooth thedata by using of least-squares fitting method.Through the study of the theory of Support Vector Machine,we identify that there is aproblem in the Support Vector Machine that it’s performance depends on the parametersetting, including penalty parameter and kernel parameters. Both Genetic Algorithms andParticle Swarm Optimization are good optimization methods.We use these two methodsrespectively for Support Vector Machine parameter optimization. In order to obtain betterclassification results, we uses Principal Component Analysis method to reduce thedimension of flight data. Through experimental verification,using Principal ComponentAnalysis to dimensionality reduction and Genetic Algorithms to select parameter canobtain the best classification.To solve the problem of identifying multi-class actions by Support Vector Machine,wecompares several commonly used multi-class Support Vector Machine,and decide to usedecision tree-support vector machines to identify the multi-class actions.Then a methodbased on experience of matching degree is proposed to construct decision-tree.Comparing“one against one” multi-class classification Support Vector Machine method and partialbinary tree multi-class classification Support Vector Machine method, the method proposedin this article is more accurate for identification in shorter time. At last,a helicopter3D reappearance system is designed by using of Multigen Creatorand Multigen Vega Prime.It enable the use of flight data to drive the3D helicopter modelflying in visual simulation scenarios.